Re: [PATCH] crypto: ccp: Fix SNP panic notifier unregistration

[Tom Lendacky <thomas.lendacky@amd.com>]

On 6/2/25 14:10, Ashish Kalra wrote:
> From: Ashish Kalra <ashish.kalra@amd.com>
> 
> Panic notifiers are invoked with RCU read lock held and when the
> SNP panic notifier tries to unregister itself from the panic
> notifier callback itself it causes a deadlock as notifier
> unregistration does RCU synchronization.

You mean that during a panic, __sev_snp_shutdown_locked() is trying to
unregister the notifier?

Wouldn't it be better to check if a panic is in progress and not try to
perform the unregister?

Or, is snp_panic_notifier() resilient enough to just always have it
registered / unregistered on module load/unload?

Also, wouldn't a better check be snp_panic_notifier.next != NULL during
sev_pci_exit()?

Thanks,
Tom

> 
> Fix SNP panic notifier to unregister itself during module unload
> if SNP is initialized.
> 
> Fixes: 19860c3274fb ("crypto: ccp - Register SNP panic notifier only if SNP is enabled")
> Signed-off-by: Ashish Kalra <ashish.kalra@amd.com>
> ---
>  drivers/crypto/ccp/sev-dev.c | 7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/crypto/ccp/sev-dev.c b/drivers/crypto/ccp/sev-dev.c
> index 8fb94c5f006a..942d93da1136 100644
> --- a/drivers/crypto/ccp/sev-dev.c
> +++ b/drivers/crypto/ccp/sev-dev.c
> @@ -1787,9 +1787,6 @@ static int __sev_snp_shutdown_locked(int *error, bool panic)
>  	sev->snp_initialized = false;
>  	dev_dbg(sev->dev, "SEV-SNP firmware shutdown\n");
>  
> -	atomic_notifier_chain_unregister(&panic_notifier_list,
> -					 &snp_panic_notifier);
> -
>  	/* Reset TMR size back to default */
>  	sev_es_tmr_size = SEV_TMR_SIZE;
>  
> @@ -2562,4 +2559,8 @@ void sev_pci_exit(void)
>  		return;
>  
>  	sev_firmware_shutdown(sev);
> +
> +	if (sev->snp_initialized)
> +		atomic_notifier_chain_unregister(&panic_notifier_list,
> +						 &snp_panic_notifier);
>  }